基于含水层DNAPL污染修复替代模型的多目标优化研究

基于Kriging方法建立表面活性剂强化修复DNAPL污染含水层的替代模型，与混合多目标算法NSGAII-HCS（Nondominated sortinggenetic algorithm II-Hill climber with step）耦合，实现修复成本最小化和治理效率最大化的多目标优化.以三维非均质承压含水层中PCE污染物的运移与修复过程为例，采用UTCHEM程序模拟表面活性剂强化修复含水层过程.将Kriging替代模型与多相流模型的输出结果进行对比，两种模型得到的含水层中PCE去除效率的平均相对拟合误差为0.80%，相关系数为0.9992，表明Kriging模型可以有效替代多相流模型.进一步将替代模型的Pareto最优解与相应的多相流模型的模拟值进行比较，得到两种模型的平均相对拟合误差仅为0.70%，相关系数达0.9998，表明在多目标优化的迭代求解过程中可以直接调用Kriging替代模型，而无须重复调用多相流模型的大负荷运算，从而为制定表面活性剂强化含水层修复决策提供一种稳定可靠的多目标优化方法.

A Kriging-based surrogate model for multi-objective optimization of DNAPL-contaminated aquifer remediation

A combined simulation-optimization model that integrates a new hybrid multi-objective genetic algorithm (Nondominated sorting genetic algorithm II-Hill climber with step, NSGAII-HCS) with a kriging surrogate model was developed for identifying the optimal designs of surfactant-enhanced aquifer remediation (SEAR) at a saturated heterogeneous aquifer site contaminated by Tetrachloroethylene (PCE). In the combined model, a three-dimensional multiphase and multicomponent compositional finite difference simulator (UTCHEM) was utilized to simulate the process of SEAR. The fitting mean relative error of removal efficiency output from the kriging-based surrogate model and the SEAR simulation model was only 0.80%, and the correlation coefficient was up to 0.9992, indicating that the surrogate model can convincingly replace the SEAR simulation model. Furthermore, the comparisons of Pareto optimal solutions based on the surrogate model and the SEAR simulation model indicated that the mean relative error of the optimal solutions and their correlation coefficient were 0.70% and 0.9998, respectively. The regression analysis results demonstrated that the proposed kriging-based surrogate models is able to predict the evolution of SEAR and the simulation-optimization tool based on the surrogate model is of lower variability and higher reliability.